From Man vs. Machine to Man + Machine

A Moment in History: Man vs. Machine

• In 1997, world champion Garry Kasparov lost to IBM’s Deep Blue.

From Adversaries to Allies

• After losing, Kasparov didn’t give up — he adapted.

• He introduced “centaur chess”, pairing human intuition with machine power.

From Chess to Finance

Kasparov’s idea of man + machine wasn’t just for chess.

This study brings that question into finance:

• Can human analysts and AI models compete — or collaborate — in forecasting the market?

• To explore this, the authors created an AI analyst and compared it to real professionals.

Meet the Analysts

The Prediction Setup

In this study, both human and AI analysts forecast 12-month-ahead stock returns using target prices.

• Target prices reflect analysts’ valuation beliefs
  
• Predicted return = (Target – Current Price) / Current Price

Research Questions

This study addresses three core questions:

1. Can an AI analyst predict 12-month-ahead stock returns more accurately than human analysts?

2. When do human analysts outperform AI — and why?

3. Does combining AI and human forecasts improve accuracy and reduce large errors?

Methodology: Overview

• The authors construct an AI analyst using machine learning
• The model forecasts 12-month-ahead stock returns derived from target prices
• Each forecast is based on a rolling 3-year window of past data, simulating real-time prediction
• Human forecasts are not used as inputs — predictions are made entirely from public information

Methodology: Modeling Approach

The AI analyst uses an ensemble of machine learning models:

• Long Short-Term Memory (LSTM) networks — capture sequential patterns in financial time series

• Random forest — robust to noise and effective with many predictors

• Gradient boosting — excels at handling non-linear relationships and variable interactions

• The ensemble leverages the strengths of each model type to improve predictive accuracy.

The Data: AI Analyst Input Overview

The AI analyst is trained on a wide variety of publicly available information.

• Human analyst data: I/B/E/S target prices (2001–2018)
• AI model: trained on 3 years of historical data prior to each forecast
• Forecast target: 12-month stock return based on target price

The Data: Categories of inputs

The model draws from six key categories of inputs:

graph TD
  A[AI Analyst Inputs] --> B[Firm Data]
  A --> C[Industry Data]
  A --> D[Macro Data]
  A --> E[Filings]
  A --> F[Sentiment]
  A --> G[Innovation]

Structure:

• Structured Inputs

• Textual and Alternative Inputs

The Data: Structured Inputs

Firm Data: - Size, book-to-market, ROA - Leverage, accruals - Past returns (1–36 months), volatility - Amihud illiquidity

Industry Data: - Fama-French industry groupings - Industry competition (text-based) - Product market fluidity

Macro Data: - GDP growth, CPI, oil prices - Treasury yields, credit spreads - Market-level illiquidity

The Data: Textual and Alternative Inputs

Filings (10-K, 10-Q, 8-K): - Sentiment and tone (Loughran–McDonald) - Readability and complexity - Text similarity and novelty

Sentiment: - RavenPack news sentiment - Twitter-based firm sentiment (Cao et al. 2021a)

Innovation: - Patent value estimates (Kogan et al. 2017)

Results: Performance – AI vs. Human

AI uses only public data, with no access to analyst forecasts or private information.

• The AI analyst outperforms human analysts in approximately 54.5% of stock forecasts.
  • This outperformance is consistent across time, sectors, and market conditions.

Results: When Humans Outperform AI

While the AI model performs better overall, human analysts outperform in specific contexts:

• Firms with high intangible assets (e.g. R&D, brand value)
• Firms facing financial distress or complex restructuring
• Stocks with low liquidity or limited trading activity
• Situations requiring qualitative judgment or soft information

Results: Man + Machine – The Hybrid Model

Combining AI forecasts with human analyst inputs creates a hybrid “centaur” analyst.

• The hybrid model slightly outperforms standalone AI in forecast accuracy.
• It improves consistency and reduces large forecast errors.
• Man + Machine captures the strengths of both:
  • AI’s scale and objectivity
  • Human analysts’ intuition and contextual judgment

The Hybrid Model Avoids Extreme Errors

• Extreme errors = forecasts in the top 10% of squared prediction errors

• Man + Machine avoids:

  • 90.7% of extreme errors made by human analysts
  • 43.6% of extreme errors made by the AI model

Conclusion

1. Can AI outperform human analysts?
   • Yes, AI beats the best human analyst in ~54.5% of cases using only public data.
2. When do human analysts outperform AI?
   • In complex settings like illiquidity, distress, and high intangibles — where qualitative judgment matters.
3. Can combining AI and human forecasts improve results?
   • Yes, the hybrid model slightly improves accuracy and greatly reduces extreme forecasting errors.